Bore Gauge MSA: Calibrated but Not Capable
A Green Belt in a bearing plant is chartered to cut bore scrap from 6.0% to 2.0%. The bore diameter specification is 25.000 ± 0.020 mm. Before the Measure phase, the team runs a crossed Gage R&R on the production bore gauge: 3 operators, 10 parts, 3 trials. The ANOVA output gives repeatability (EV) 0.0045 mm, reproducibility (AV) 0.0032 mm, part variation (PV) 0.0120 mm, and total GRR 0.0055 mm. All 10 parts were pulled from a single hour of production, so they represent the current, already-narrow process. Operator B reads every part about 0.004 mm lower than operators A and C. No study part is outside 25.020 mm, but two parts sit within 0.003 mm of that limit. The gauge is calibrated and the master-ring check is inside tolerance. The quality manager says: 'The gauge is fine - it is calibrated and it matches the master. Skip the MSA and start the Improve phase; we are behind schedule.'
1. Evaluate whether this measurement system is fit for the project's purpose and for the ongoing SPC chart the team plans to use. Show the metrics you use to decide.
Read the worked answer
Compute the three standard metrics rather than relying on the calibration certificate.
- %GRR of tolerance = 0.0055 / 0.040 = 13.8% - this looks acceptable, but it only answers 'can the gauge sort good from bad parts?'
- %GRR of study variation = 0.0055 / √(0.0055² + 0.0120²) = 0.0055 / 0.0132 = 41.7% - far above the 30% limit.
- ndc = 1.41 × (PV/GRR) = 1.41 × 2.18 = 3.08 → 3 distinct categories; at least 5 is needed.
Interpretation: the gauge is marginal for product acceptance but not fit for process control or for proving a variation reduction. With ndc = 3, a real improvement in bore spread can be hidden inside gauge noise, so the team could 'improve' the process and see nothing on the chart - or see a signal that is pure appraiser bias. The 0.004 mm operator offset on B is roughly 20% of the tolerance and about 73% of the total GRR, so reproducibility is the dominant component, not the instrument itself. Calibration proves the gauge agrees with a master at one point; it says nothing about discrimination across the range the process actually uses.
2. The quality manager wants to move straight to Improve. What is your response, and what specific corrective actions do you take before re-running the study?
Read the worked answer
The MSA cannot be waived, but the fix is smaller than it first appears. The GRR is inflated by two correctable problems:
- Appraiser effect. Watch Operator B measure; check seating depth, gauge orientation, and whether B uses the same fixturing stop as A and C. Retrain to a documented method and re-check bias on a known master. If bias persists after retraining, recalibrate or reassign B.
- Part selection. Parts came from one hour, so PV is understated relative to the true process spread? No - the opposite risk here: the parts span only part of the process, which can distort the ratio in either direction. Reselect 10 parts covering the full operating range, including parts deliberately near the 25.020 mm limit, because that is where the team must be able to decide pass/fail.
Re-run the crossed study. Accept the system only when %GRR of study variation is under 30% and ndc ≥ 5; for scrap decisions at the limit, aim under 10% of tolerance. Only then freeze the baseline scrap rate and start the SPC chart. Going to Improve now would make the 6.0% → 2.0% goal unverifiable: you would be spending project time on a gauge that cannot tell you whether you succeeded.